如何以每个记录与“上一个”记录连接的方式自行连接表?

时间:2013-03-20 15:11:11

标签: sql sql-server performance sql-server-2008

我有一个MS SQL表,其中包含具有以下列的库存数据:Id, Symbol, Date, Open, High, Low, Close

我想自行加入表格,因此我可以获得Close的日常变化百分比。

我必须创建一个将表连接到自身的查询,每个记录都包含来自前一个会话的数据(请注意,我不能使用昨天的日期)。

我的想法是做这样的事情:

select * from quotes t1
inner join quotes t2
on t1.symbol = t2.symbol and
t2.date = (select max(date) from quotes where symbol = t1.symbol and date < t1.date)

但我不知道这是否是正确/最快的方式。在考虑性能时应该考虑什么? (例如,将UNIQUE索引放在(符号,日期)对上可以提高性能吗?)

此表中每年将有大约100,000条新记录。我正在使用MS SQL Server 2008

7 个答案:

答案 0 :(得分:9)

一种选择是使用递归cte(如果我正确理解你的要求):

WITH RNCTE AS (
  SELECT *, ROW_NUMBER() OVER (PARTITION BY symbol ORDER BY date) rn
        FROM quotes
  ),
CTE AS (
  SELECT symbol, date, rn, cast(0 as decimal(10,2)) perc, closed
  FROM RNCTE
  WHERE rn = 1
  UNION ALL
  SELECT r.symbol, r.date, r.rn, cast(c.closed/r.closed as decimal(10,2)) perc, r.closed
  FROM CTE c 
    JOIN RNCTE r on c.symbol = r.symbol AND c.rn+1 = r.rn
  )
SELECT * FROM CTE
ORDER BY symbol, date

SQL Fiddle Demo

如果您需要将每个符号的运行总计用作百分比更改,那么很容易为该数量添加额外的列 - 不能完全确定您的意图是什么,所以上面只是将当前关闭以前的结算金额。

答案 1 :(得分:3)

像这样的东西在SQLite中工作:

SELECT ..
FROM quotes t1, quotes t2
WHERE t1.symbol = t2.symbol
    AND t1.date < t2.date
GROUP BY t2.ID
    HAVING t2.date = MIN(t2.date)

鉴于SQLite是最简单的一种,也许在MSSQL中,这也可以在最小的变化下工作。

答案 2 :(得分:1)

你做这样的事情:

with OrderedQuotes as
(
    select 
        row_number() over(order by Symbol, Date) RowNum, 
        ID, 
        Symbol, 
        Date, 
        Open, 
        High, 
        Low, 
        Close
      from Quotes
)
select
    a.Symbol,
    a.Date,
    a.Open,
    a.High,
    a.Low,
    a.Close,
    a.Date PrevDate,
    a.Open PrevOpen,
    a.High PrevHigh,
    a.Low PrevLow,
    a.Close PrevClose,

    b.Close-a.Close/a.Close PctChange

  from OrderedQuotes a
  join OrderedQuotes b on a.Symbol = b.Symbol and a.RowNum = b.RowNum + 1

如果您将最后一个连接更改为左连接,则每个符号的第一个日期都会有一行,不确定是否需要。

答案 3 :(得分:1)

(symbol, date)

上的索引
SELECT *
FROM quotes q_curr
CROSS APPLY (
  SELECT TOP(1) *
  FROM quotes
  WHERE symbol = q_curr.symbol
    AND date < q_curr.date
  ORDER BY date DESC
) q_prev

答案 4 :(得分:0)

你可以这样做:

DECLARE @Today DATETIME
SELECT @Today = DATEADD(DAY, 0, DATEDIFF(DAY, 0, CURRENT_TIMESTAMP))

;WITH today AS
(
    SELECT  Id ,
            Symbol ,
            Date ,
            [OPEN] ,
            High ,
            LOW ,
            [CLOSE],
            DATEADD(DAY, -1, Date) AS yesterday 
    FROM quotes
    WHERE date = @today
)
SELECT *
FROM today
LEFT JOIN quotes yesterday ON today.Symbol = yesterday.Symbol
    AND today.yesterday = yesterday.Date

这样就可以限制“今天”的结果,如果这是一个选项。

编辑:作为其他问题列出的CTE可能运行良好,但在处理100K行或更多行时,我倾向于使用ROW_NUMBER犹豫不决。如果前一天可能并不总是昨天,我倾向于在自己的查询中拉出前一天的支票,然后将其用作参考:

DECLARE @Today DATETIME, @PreviousDay DATETIME
SELECT @Today = DATEADD(DAY, 0, DATEDIFF(DAY, 0, CURRENT_TIMESTAMP));
SELECT @PreviousDay = MAX(Date) FROM quotes  WHERE Date < @Today;
WITH today AS
(
    SELECT  Id ,
            Symbol ,
            Date ,
            [OPEN] ,
            High ,
            LOW ,
            [CLOSE]
    FROM quotes 
    WHERE date = @today
)
SELECT *
FROM today
LEFT JOIN quotes AS previousday
    ON today.Symbol = previousday.Symbol
    AND previousday.Date = @PreviousDay

答案 5 :(得分:0)

你有什么好。我不知道将子查询翻译成连接是否有帮助。但是,你要求它,所以这样做的方法可能是再次将表连接到自己。

select *
from quotes t1
inner join quotes t2
   on t1.symbol = t2.symbol and t1.date > t2.date
left outer join quotes t3
   on t2.symbol = t3.symbol and t2.date > t3.date
where t3.date is null

答案 6 :(得分:0)

您可以使用CTEROW_NUMBER排名功能

选项
 ;WITH cte AS
 (
  SELECT symbol, date, [Open], [High], [Low], [Close],
         ROW_NUMBER() OVER(PARTITION BY symbol ORDER BY date) AS Id
  FROM quotes
  )
  SELECT c1.Id, c1.symbol, c1.date, c1.[Open], c1.[High], c1.[Low], c1.[Close], 
         ISNULL(c2.[Close] / c1.[Close], 0) AS perc
  FROM cte c1 LEFT JOIN cte c2 ON c1.symbol = c2.symbol AND c1.Id = c2.Id + 1
  ORDER BY c1.symbol, c1.date

为了提高性能(避免排序和RID查找),请使用此索引

CREATE INDEX ix_symbol$date_quotes ON quotes(symbol, date) INCLUDE([Open], [High], [Low], [Close])

SQLFiddle

上的简单演示